Ennyd does this on your Mac: it transcribes your local videos (no upload, any size), you pick passages right from the transcript, and it exports them as clips. Fast export is lossless, with no re-encode. 48-second demo.

Reliable Testing by Agents - Best Practices : 1. Start from intent, not the implementation Write down what the user should be able to rely on before reading the code under test. A test derived from the code tends to check what the code does, bugs included. 2. Assert outcomes someone relies on Check the result the user would notice, not that some code ran. "The job finished" can pass while the output is empty. 3. Test where bugs live The happy path is rarely where things break. Choose the cases that matter for this behavior, and say why you skipped the rest. 4. Prove every test can fail A test you have never seen fail proves nothing. It might click the wrong thing, assert something that is always true, or pass for an unrelated reason. Break the behavior on purpose and watch the test catch it. 5. Verify independently The author of the code should not be its only verifier. An agent is especially prone to confirming its own work, because it already believes the result. 6. Triage failures instead of rerunning to green Every failure has a cause. Rerunning until it passes hides the cause and teaches you to ignore the test. 7. Report evidence honestly A summary is not evidence. Back every claim of "done," "passing" or "safe" with something a reader can check.

It is our mouth, throat, and tongue that limit our thought capacity. Give it to the eyes, instead.. and Sense in Spatial.

Some principles of Agentic Engineering, in Enterprise environments. 1. The product is the work. The platform is plumbing. 2. Guard the one-way doors and nothing else. 3. Speed is a safety feature. 4. Rollback is cheaper than prevention. 5. Fix the cause, never guard the guard. 6. Failures resume, they don't restart. 7. One way to do each thing. 8. Test what users touch. 9. Trust people and judgment over layers. 10. Every rule pays rent. The short version: guard the data and the doors that can't be undone, keep everything else fast and simple, and when in doubt leave it out.

dependency management will be a full time job soon

Instead of: "I want a script that loads data from an API into a database, but make sure if it crashes halfway through it doesn't duplicate the rows when I run it again." Say: "Write an idempotent Python script to ingest API data into PostgreSQL, ensuring it gracefully handles failures without violating our Single Source of Truth." Instead of: "We have a rule that developers can't push code directly to production without a manager looking at it first, but we want it automated." Say: "Design a CI/CD workflow that enforces separation of duties and policy as code before promotion to production."

Using precise vocabulary is the highest-leverage way to communicate with LLMs. When you use industry terms of art, you bypass paragraphs of explanation because the LLM instantly recognizes the underlying design patterns, constraints, and historical contexts associated with those exact phrases.

Platform Engineering Terms of Art : The road itself : - Paved road / golden path: the one supported, safe way to do a common task. - Self-service: people do routine tasks themselves without waiting on an expert. - Internal developer platform (IDP): the shared set of tools, templates and workflows. - Toil: repetitive manual work that adds no lasting value and should be automated. - Idempotent: safe to run twice; the second run changes nothing. - Declarative vs imperative: describing the desired end state vs scripting each step. Identity and provenance : - Immutable artifact: built once, never modified, identified by a content hash. - Build once, promote many: the exact same artifact moves through every environment. - Provenance / attestation: verifiable proof of how and from what an artifact was built. - Software supply chain, SLSA, SBOM: securing and listing everything that goes into a build. - Pinning / lockfiles / reproducible builds: exact dependency versions, so rebuilds match. Safety and gates : - Policy as code: rules enforced automatically by tooling rather than by checklists. - Guardrails vs gates: guardrails keep work on track; gates pause for a decision. - One-way door / two-way door: irreversible vs easily reversible decisions. - Fail closed: when uncertain, stop rather than proceed. - Least privilege / short-lived credentials: minimal access that expires quickly. - Separation of duties: the person or system that makes a change doesn't approve it. Release practice : - Continuous integration / continuous delivery (CI/CD): frequent merging and always-releasable code. - Staging / environment parity: pre-production environments that mirror production. - Environment drift: environments slowly and silently diverging. - Blue-green, canary, progressive delivery: switching or ramping traffic to a new version gradually. - Feature flags: turning functionality on or off without redeploying. - Rollback vs roll-forward: reverting to the last good version vs fixing forward. - Change freeze / release window: periods when changes are restricted or allowed. Reliability and recovery : - SLO / SLI / error budget: a reliability target, its measurement, and the failure allowed. - RPO / RTO: maximum acceptable data loss, and maximum acceptable recovery time. - Backup verification / restore drills: a backup counts only once it has been restored. - Disaster recovery (DR): recovering from losing a whole system or site. - Runbook / playbook: written procedures for routine operations and incidents. - Observability vs monitoring: understanding system behavior vs alerting on known conditions. - Freshness check / dead man's switch: avent didn't happen. Learning and improvement - Blameless postmortem: analyzing failures in the system, not blaming people. - Friction log: a running record of rough edges to remove. - Paving the path: turning a manual procedure into one automated command. - DORA metrics: deployment frequency, lead time for changes, change failure rate, time to restore. - Platform as a product: treating internal tooling as a product with users and feedback.

There are vibe coded apps, which are apps made by non developers. There are AI assisted apps, which are apps made by developers.